Article:

Journal Version: Zijian Zhao, Zhijie Cai, Tingwei Chen, Xiaoyang Li, Hang Li, Qimei Chen, Guangxu Zhu*, "KNN-MMD: Cross Domain Wireless Sensing via Local Distribution Alignment "(under review, IEEE Transactions on Mobile Computing (TMC))

Conference Version: Zijian Zhao, Zhijie Cai, Tingwei Chen, Xiaoyang Li, Hang Li, Qimei Chen, Guangxu Zhu*, "Does MMD Really Align? A Cross Domain Wireless Sensing Method via Local Distribution", IEEE/CIC ICCC 2025 Patent: 赵子健, 朱光旭, 陈琪美, 韩凯峰 "基于少样本学习的模型对对象识别的方法及相关设备"（专利号：ZL202411074110，2024）.

Notice: We have uploaded our dataset (RS2002/WiFall · Datasets at Hugging Face) to Hugging Face.

Public Dataset: WiGesture

Proposed Dataset: WiFall (./WiFall)

Refer to RS2002/CSI-BERT: Official Repository for The Paper, Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing (github.com)

To run the model, follow these instructions based on the dataset you are using. For the WiGesture Dataset, use the train.py script, and for the WiFall Dataset, use the train_fall.py script. The steps to execute them are the same, and here we provide an example using train.py.

python train.py --k <shot number> --n <neighbor number for KNN> --p <select the top p samples from testing set for MK-MMD (p<1)> --task <action or people> --lr <learning rate>

Make sure to replace the following placeholders with the appropriate values:

• <shot number>: Specify the shot number.
• <neighbor number for KNN>: Specify the number of neighbors for KNN.
• <select the top p samples from testing set for MK-MMD (p<1)>: Specify the value for p (selecting the top p samples from the testing set for MK-MMD). Note that p should be less than 1.
• <action or people>: Specify the task name as either "action" or "people".
• <learning rate>: Specify the desired learning rate.

Once you have set the appropriate values, run the command in your terminal to start the training process.

@misc{zhao2025knnmmdcrossdomainwireless,
      title={KNN-MMD: Cross Domain Wireless Sensing via Local Distribution Alignment}, 
      author={Zijian Zhao and Zhijie Cai and Tingwei Chen and Xiaoyang Li and Hang Li and Qimei Chen and Guangxu Zhu},
      year={2025},
      eprint={2412.04783},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.04783}, 
}

@INPROCEEDINGS{11149311,
  author={Zhao, Zijian and Cai, Zhijie and Chen, Tingwei and Li, Xiaoyang and Li, Hang and Chen, Qimei and Zhu, Guangxu},
  booktitle={2025 IEEE/CIC International Conference on Communications in China (ICCC)}, 
  title={Does MMD Really Align? A Cross Domain Wireless Sensing Method via Local Distribution}, 
  year={2025},
  volume={},
  number={},
  pages={1-6},
  keywords={Wireless communication;Training;Wireless sensor networks;Codes;Sensitivity;Gesture recognition;Nearest neighbor methods;Stability analysis;Sensors;Wireless fidelity;Few-shot Learning;K-Nearest Neighbors;Maximum Mean Discrepancy;Cross-domain Wireless Sensing;Channel Statement Information},
  doi={10.1109/ICCC65529.2025.11149311}}